public Neat()
{
Neural = new NeuralParameters();
Mutation = new MutationParameters();
Speciation = new SpeciationParameters();
Reproduction = new ReproductionParameters();
Structure = new StructureParameters();
}
public XorNetworkSearch(
PopulationParameters populationParameters,
NetworkParameters networkParameters,
ReproductionParameters reproductionParameters)
{
_populationParameters = populationParameters;
_neatPopulation = new Population(networkParameters, reproductionParameters);
_genomes = Enumerable.Range(0, _populationParameters.PopulationSize)
.AsParallel()
.Select(_ => _neatPopulation.CreateInitialGenome()).
ToList(_populationParameters.PopulationSize);
_archiveSize = _populationParameters.PopulationSize / 30;
_archive = new List <ParetoFrontPoint>(_archiveSize);
_isRecurrent = networkParameters.IsRecurrent;
}
public void NetworkReset()
{
var initialParameters = new NetworkParameters(10, 5, NetworkType.FeedForward)
{
InitialConnectionDensity = 1f
};
var reproductionParameters = new ReproductionParameters()
{
WeightMutations = new WeightMutations
{
OverallRouletteWheelShare = 0
},
AddConnectionRouletteWheelShare = 2f,
RemoveConnectionRouletteWheelShare = 1f,
SplitConnectionRouletteWheelShare = 1f
};
var neatPopulation = new Population(initialParameters, reproductionParameters);
var genomes = Enumerable.Range(0, 100).Select(_ => neatPopulation.CreateInitialGenome()).ToArray();
for (var i = 0; i < 500; i++)
{
var sensorValues = Enumerable.Range(0, initialParameters.SensorCount)
.Select(_ => (float)new Random().NextDouble()).ToArray();
foreach (var genome in genomes)
{
var network = genome.Network;
for (var j = 0; j < initialParameters.SensorCount; j++)
{
network.Sensors[j] = sensorValues[j];
}
network.Activate();
var effectors = network.Effectors.ToArray();
network.Activate();
Assert.Equal(effectors, network.Effectors);
}
genomes = genomes.Select(g => neatPopulation.Replicator.Reproduce(g)).ToArray();
}
}
public void AsexualReproduction()
{
var initialParameters = new NetworkParameters(2, 3, NetworkType.FeedForward)
{
InitialConnectionDensity = 0.5f
};
var reproductionParameters = new ReproductionParameters()
{
WeightMutations = new WeightMutations
{
OverallRouletteWheelShare = 50
},
AddConnectionRouletteWheelShare = 5f,
RemoveConnectionRouletteWheelShare = 5f,
SplitConnectionRouletteWheelShare = 5f
};
var neatPopulation = new Population(initialParameters, reproductionParameters);
var sensorValues = Enumerable.Range(0, initialParameters.SensorCount)
.Select(_ => (float)new Random().NextDouble()).ToArray();
var genome = neatPopulation.CreateInitialGenome();
var network = genome.Network;
for (var i = 0; i < initialParameters.SensorCount; i++)
{
network.Sensors[i] = sensorValues[i];
}
network.Activate();
for (var i = 0; i < 50000; i++)
{
genome = neatPopulation.Replicator.Reproduce(genome);
var newNetwork = genome.Network;
for (var s = 0; s < initialParameters.SensorCount; s++)
{
newNetwork.Sensors[s] = sensorValues[s];
}
newNetwork.Activate();
}
}
public SplitConnectionWeightsSearch(
PopulationParameters populationParameters,
NetworkParameters networkParameters,
ReproductionParameters reproductionParameters)
{
_populationParameters = populationParameters;
_neatPopulation = new Population(networkParameters, reproductionParameters, false);
_genomes = Range(0, _populationParameters.PopulationSize)
.Select(_ => _neatPopulation.CreateInitialGenome());
_archiveSize = _populationParameters.PopulationSize / 2;
_archive = new List <ParetoFrontPoint>(_archiveSize);
WeightSamples = Range(0, 300)
.Select(_ => RandomSource.Range(0, MaxAbsWeightValue))
.ToArray();
InputSamples = Range(0, 60)
.Select(_ => RandomSource.Range(0, 1))
.ToArray();
}
